WWDC-MLX

Overview

Code Walkthrough for Real-World Applications of MLX in WWDC OMT 2025. We use the local embedding model .bge_small to create an on-device local semantic search index, pulling in from the latest Apple WWDC 2025 documentation using local embeddings and a fast in-memory search index. The app crawls the official Apple documentation, generates vector embeddings for the content, and enables natural language search with similarity ranking.

Features

• Semantic Search: Enter a query and find the most relevant documentation snippets using vector similarity.
• Automatic Crawling: The app fetches and indexes content from the official WWDC 2025 documentation.
• Modern SwiftUI Interface: Clean, responsive UI with search bar and result list.
• Source Linking: Each result links directly to the original Apple documentation page.
• Local Embedding Generation: Uses MLX and MLXEmbedders for on-device embedding computation.

Dependencies

This project uses Swift Package Manager and includes the following dependencies (see Package.resolved for exact versions):

• MLX Swift
• MLX Swift Examples
• Swift Transformers
• SwiftSoup

Getting Started

1. Clone the repository:

   git clone https://github.com/mcantillon21/local-search.git
   cd WWDC-MLX

2. Open in Xcode:

   • Open WWDC-MLX.xcodeproj in Xcode (version 15 or later recommended).

3. Build and Run:

   • Select the WWDC-MLX scheme and run the app on your Mac.

4. Search:

   • Wait for the initial indexing to complete, then enter your search query in the search bar.

Notes

• The app requires network access to crawl and fetch documentation. But once indexed, works without internet.
• Note: Embedding generation and crawling may take some time on first launch.
• This is just an unoptimized demo, but you could replicate this process for any website, any of your data.